⊂The Heterogenous Cost of Wage Rigidity: Evidence
and Theory∗
Ester FaiaGoethe University Frankfurt and CEPR
Vincenzo PezoneGoethe University Frankfurt and SAFE
Current draft: January 2019.
Abstract
Using a unique confidential contract level dataset merged with firm-level asset price data,we find robust evidence that firms’ stock market valuations and employment levels respondmore to monetary policy announcements the higher the degree of wage rigidity. Data on therenegotiations of collective bargaining agreements allow us to construct an exogenous measureof wage rigidity. The amplification induced by wage rigidity is stronger for firms with high laborintensity and low profitability. There are clear distributional consequences of monetary policy.We rationalize the evidence through a model in which firms in different sectors feature differentdegrees of wage rigidity due to staggered renegotiations vis-a-vis unions.
JEL:Keywords: heterogeneous monetary policy response, distributional consequences of monetary
policy, employer-employee level dataset, monetary policy surprise shocks, heterogeneous wagerigidity.
∗We gratefully acknowledge financial support from SAFE LOEWE center and data access from the IstitutoNazionale della Previdenza Sociale (INPS) through the VisitINPS Scholars Program. We thank Tito Boeri, EdoardoDi Porto, Pietro Garibaldi, Marcin Kacperczyk, Leonid Kogan, Matthias Meier (discussant), Marco Pagano, PaoloNaticchioni, Javier Ruiz (discussant) Michael Weber and participants at various conferences and seminars, and NanHu and Soren Karau for excellent data research assistance on the data. The views expressed in the articles are thoseof the authors and do not involve the responsibility of INPS. Correspondence to: Ester Faia, Chair in Monetaryand Fiscal Policy, Goethe University Frankfurt, Theodor W. Adorno Platz 3, 60323, Frankfurt am Main, Germany.E-mail: [email protected]. Webpage: http://www.wiwi.uni-frankfurt.de/profs/faia/.
1
1 Introduction
The need to understand the distributional consequences of monetary policy has been growing
rapidly. Very recently the attention has shifted toward firms’ heterogeneity. There has been in-
creasing evidence that heterogeneity in the degree of price stickiness and in financial conditions1
plays a significant role in the transmission of monetary policy shocks to firms’ stock market valu-
ations or investment, and some authors have built models to rationalize those channels. Despite
the fact that labor market conditions play a key role in the transmission of monetary policy and
that wage rigidity have the longest tenure in macroeconomics as driver of monetary non-neutrality,
at least since the seminal work by Fischer (1977) and Taylor (1979), there has been little work
on the heterogenous consequences of this channel. The question is meaningful also since a recent
literature has emphasized that there is substantial heterogeneity in firm-level employment dynam-
ics that is masked by aggregate data2. A major obstacle to such a study has been the absence
of reliable contract-level data on wages and wage renegotiations dates, something that lead many
macroeconomists to doubt the relevance of wage rigidity.
Leveraging on a unique micro level dataset on collective bargaining agreements, which com-
prises the entire Italian industrial landscape, merged with high frequency data on stock prices for
individual listed firms and other variables for firms’ economic activity, such as employment levels,
we can answer the question above. The Italian labor market offers a unique environment as vir-
tually all workers are covered by collective bargaining agreements. Contract renegotiations occur
at predetermined schedules and firms in different sectors have different renegotiations dates. This
staggered structure, coupled with the large firm-level variation, allows us to construct a measure of
wage rigidity based solely on the timing of collective agreements. Our proxy exploits information
on the time left before the expiration of the agreement in force on the day of a monetary policy
1For the role of nominal rigidities, see Nakamura and Steinsson (2008), Gorodnichenko and Weber (2016), andCarvalho (2006). For the importance of heterogeneity in financial conditions, see Bahaj, Foulis, Pinter and Surico(2018), Jeenas (2018) and Ottonello and Winberry (2018).
2Davis, Haltiwanger, and Schuh 1996; Fort, Haltiwanger, Jarmin, and Miranda 2013; Dinlersoz, Kalemli-Ozcan,Hyatt, and Penciakova 2018.
announcement. The further the expiration day, the longer the firm will be unable to adjust its
wages to the new economic environment. Hence, higher distance to contract renewal will amplify
the impact of a monetary policy shock. Collective agreements are outside of an individual firm’s
control. For this reason, our proxy can be considered an exogenous measure of nominal rigidities,
also in comparison with others based on firms’ price level data. This is a key novelty of our analysis.
Our empirical methodology connects changes in firms’ stock market values and employment
to monetary policy surprises. We identify monetary policy shocks at high frequency via changes in
swap rates on money market rates, namely the main ECB policy rate, shortly after and before the
time an ECB meeting takes place3.
We uncover significant and robust heterogenous responses of firms’ stock valuations and em-
ployment to high-frequency identified monetary policy surprises. Firms with a more rigid labour
costs structure, as implied by our proxy, induce higher responsiveness in their stock market valua-
tions. The highly refined information available in our comprehensive matched employer-employee
dataset, allows us also to condition to several firms and worker characteristics, disentangling various
aspects of the labour demand channel. On the firm side we find that the amplification induced by
wage rigidity is stronger for firms with high labor intensity, low profitability, more flexible staffing
arrangements, and during times of economic downturn. On the workers’ side we find that long-term
workers are fairly unaffected by the combined effects of monetary policy shocks and wage rigidity,
suggesting that most of the burden of such adjustments is borne by short-term workers.
We rationalize this evidence by constructing a novel macro model whereby firms in different
sectors face different degrees of wage rigidity due to staggered renegotiations with unions4. Firms
in the model are monopolistic, so that fluctuations in marginal costs translate into fluctuations of
future discounted profits, hence firms’ value. Using simulated data from our model we replicate
3Early work on high frequency identification of monetary policy shocks includes Gurkaynak, Sack, and Swanson(2005) and Barakchian and Crowe (2013). Following a recent approach by Corsetti, Duarte, and Mann (2018) weemploy swap rates to compute the surprise component.
4This structure captures well the realm of many Western European countries (Ronchi and di Mauro (2017)),whereby contracts are renegotiated at fixed length, but staggering occurs among firms in different sectors.
3
the regressions done in the empirical part. The model-based regressions mirror their empirical
counterparts. The model transmission mechanism runs as follows. Firms facing larger distance
between contract renegotiation and the monetary policy shock cannot re-adjust marginal costs.
This induces larger fluctuations in employment demands and in mark-ups. The latter in turn
translates into larger fluctuations of future discounted profits or valuations.
Our paper makes three contributions. First, we construct a truly exogenous measure of nominal
rigidity at the firm level, based on the timing of wage contract renewals. Second, we use this proxy
to uncover heterogenous responses of firms, in terms of stock values and employment, to high-
frequency identified monetary policy shocks. The heterogeneity in the response also varies with a
number of firms’ and workers’ characteristics. Third, we construct a dynamic general equilibrium
with sectorial firms’ heterogeneity in the degree of wage rigidity. The model replicates well the
empirical results.
The rest of the paper proceeds as follows. Section 2 compares our work to past literature.
Section 3 presents a simple model to assess the role of monetary policy in the presence of rigid
wages. The model also serves the purpose of providing guidance to the subsequent econometric
specifications. Section 4 provides some institutional background and describes our data. Section 5
presents our empirical evidence. Section 6 develops a fully dynamic multi-sector general equilibrium
model that can rationalize and replicate the evidence. Section 7 concludes. Tables and figures
follow.
2 Comparison to the Literature
Our paper is first and foremost related to the recent literature assessing the heterogenous effects
of monetary policy announcements on firms’ activity and stock market values. The literature
examined mainly two forms of heterogeneity, in price stickiness (Gorodnichenko and Weber (2016)
and Carvalho (2006))5 and in financial conditions (Bahaj, Foulis, Pinter, and Surico (2017), Jeenas
5Evidence on the importance of price stickiness heterogeneity is also in Nakamura and Steinsson (2008), Zbaracki,Ritson, Levy, Dutta, and Bergen (2004), Anderson, Jaimovich, and Simester (2015) and Blinder (1991).
4
(2018), and Ottonello and Winberry (2018)). Despite the importance of labour market frictions
and wage rigidities in particular, the literature has largely neglected them so far.
Our empirical methodology builds and departs from Gorodnichenko and Weber (2016). They
merge confidential micro-level data underlying the producer price index (PPI) with stock price data
of individual firms from NYSE Trade and Quote (TAQ) and study how stock returns of firms with
different frequencies of price adjustment respond to high-frequency identified monetary shocks.
Contrary to them focus on wage rigidity; while the frequency of price adjustment might be, in
some respects, a choice variable for a firm’s managers, the factors affecting the timing of the wage
renegotiations in our context are largely outside of a firm’s control. Hence, our strategy exploits
arguably exogenous variation in nominal rigidities. Moreover to highlight the role of the labour
market channel we also examine the differential impact of monetary policy on firms’ employment
demand, conditioning also on various firms’ and workers’ characteristics. The latter allows us to
best exploit the distributional consequences of monetary policy.
The precise measurement of wage rigidity has traditionally been an unreachable task. Most
of past studies attempting to measure it only relied on survey data, whereby measurement errors
and large imputations vastly challenge the precision6. Our measure is instead based on direct
observation of contracts available at employer-employee matched dataset7, hence it provides the
maximum degrees of precision.
The monetary transmission mechanism in presence of wage rigidity has been studied by Olivei
and Tenreyro (2007), who use VAR evidence with aggregate data. Given our access to refined
and precise micro data, we can focus on the more novel cross-sectional effects, rather than on the
aggregate economy. Moreover, our approach, not relying on parametric assumptions, is essentially
6For the US, McLaughlin (1994) and Kahn (1997) use survey data. For Europe survey evidence is in Smith(2000) and is summarized in Kramarz (2001). The issue of measurement error in labour surveys is addressed inCard and Hyslop (1997) and Altonji and Devereux (2000). Groshen and Schweitzer (1999) use employer data, whileBewley (1998) conducted interview himself. Finally, the International Wage Flexibility Project (IWFP), whose resultsaree summarized in Dickens, Goette, Groshen, Holden, Messina, Schweitzer, Turunen, and Ward (2007), analyzesindividuals’ earnings in 31 different datasets from sixteen countries.
7See also Abowd, Corbel, and Kramarz (1999) for a study constructing measures of wage rigidity using an employer-employee matched dataset.
5
model-free.8
This paper is laterally related to a growing literature linking finance and labour market fric-
tions. Serfling (2016) and Simintzi, Vig, and Volpin (2014) show that firms reduce leverage in
the presence of higher firing costs, which imply an increase in operating leverage. In our setting,
instead, the rigidity affecting firm value and employment is given by the inability to adjust wages
following aggregate shocks.
3 A Simple Model to Guide Intuition
The effects of monetary policy are surely best appreciated in a fully dynamic quantitative model.
It is instructive however to outline the functioning of a simple model that can capture the economic
intuition behind our empirical analysis and can guide its specification. The main goal is to establish
a link between shocks to profitability and changes in firms’ market values, and to show how this
connection can be amplified by wage rigidity. We impose at this stage a minimal set of assumptions,
so that the resulting testable implications from the model can be subject to an agnostic test. Our
simple model builds on Fischer (1977). We postpone to Section 6 the presentation of a more
rigorous model that, under some parameter calibrations, can produce results that are not too far
from our reduced-form estimates.
Consider a firm whose expected per-period cash flow is equal to E[πt] = π. The firm’s horizon is
infinite, and discounts each period at a rate δ. Every two periods, the trade union representing the
workforce bargains with firms’ managers on behalf of workers. The firm’s total surplus, represented
by its cash flow, is then split with bargaining weights β and 1−β between workers and shareholders,
respectively. At the beginning of period t, all players learn the value of πt. Before the actual value
realization, the trade union and the firms’ managers agree on the total wage bill w that will accrue
to workers in period t and t+1. Since the value of πt+1 is unknown at time t, workers and managers
8Bjorklund, Carlsson, and Nordstrom Skans (2019) also construct proxies for wage rigidities based on renegoti-ation periods using Swedish administrative data. Their focus is however on aggregate consequences, rather thandistributional effects of shocks.
6
set wages wt+1 = βπ, which implies that shareholders get (1− β)πt in periods t and πt+1 − βπ in
period t+ 1.
Now consider a shock, such as a monetary policy announcement, that temporarily changes the
cash flow of the firm by an amount ∆, but only in period τ , so that πτ = π + ∆. The impact of
this shock to the firm’s value depends upon the timing of the contract’s renewal. If t = τ , meaning
that the renegotiation of the wage contract occurs immediately after the shock is realized, the firm
value is:
Vτ,τ=t = (1− β)
(π
1− δ+ ∆
)(1)
On the other hand, if the shock occurs after the renegotiation has taken place, hence if τ = t+1,
the previously bargained profit share does not reflect the changed profitability of the firm. Hence,
in this case, firm value is:
Vτ,τ=t+1 = (1− β)π
1− δ+ ∆ (2)
The difference between these two values is Vτ,τ=t+1 − Vτ,τ=t = β∆. Two immediately testable
implications emerge. First, the change in firms’ value is directly related to the distance from the
contract renegotiation. Second, the change is larger as β increases. Workers’ bargaining power
captures unions’ strength, which in turn is related to the firms’ labour share.
A few considerations are worth at this stage. First, certain indexing arrangements might
restore money neutrality; for example, the effects of wage stickiness could indeed be eliminated by
state contingent contracts. However, as we will discuss in Sections 4.1 and 4.4, wages reflect only
in a limited way changes in profitability, whereas indexation to the broad consumer price index has
been repealed in 1994 (see Manacorda (2004) for details). Collective agreements require changes
in minimum wages to occur according to a predetermined schedule, reflecting expected inflation.
However, since such changes are fully anticipated, they will be fully incorporated into stock prices
at the moment of a shock. In both those cases the model predictions will go through.
7
Second, without making strong assumptions about a firm’s profit function and its optimal price
level, it is generally impossible to tell whether a monetary policy announcement has a positive or
negative effect on its profit, hence on whether ∆ > 0 or ∆ < 0). 9 Under nominal rigidities
firms’ prices might be above or below the optimal price at the time of the announcement, see in
Figure 1. In response to monetary shocks that change firms optimal price firms’ profits might
raise or fall depending on whether the initial price level was above or below the new optimal
price. Generally speaking all effects linked to non-linear profits can be accommodated by including
the square terms of the monetary policy shock and of the firm’s stock return in the econometric
specification. Such assumptions also make the identification procedure independent from specific
models’ assumptions.10
4 Institutional Background and Data
4.1 Institutional Background
Italy has a centralized wage bargaining process, similar to other developed economies in Western
Europe, such as France, Belgium, Portugal,Finland, Iceland, and Slovenia . In other countries,
such as Austria, Denmark, Germany, Greece, Ireland, Portugal, Spain and Sweden, centralised
wage-setting is also prevalent, but more room is generally given to firm-level bargaining (See Boeri,
Ichino, Moretti, and Posch (2017)). Several characteristics are worth noticing. First, for each job
“category”, collective bargaining agreements regulate salary conditions, days of vacation, compen-
sation for extra hours, and a number of other aspects of the employee-employer relationship. While
there is a loose relationship between job categories and economic sectors, mapping each agreement
to a single industry is generally not straightforward.11 Second, collective agreements, typically valid
for two or three years, are done between unions and industry representatives, but are valid erga
9See also Gorodnichenko and Weber (2016).10Parametric approaches are followed, for example, in Olivei and Tenreyro (2007) and Bjorklund et al. (2019).11In some cases, collective agreements may regulate the labor conditions of workers with different positions within
the same sector. For example, different collective agreements are in force for banking employees and banking managers.In other cases, workers in the same industry may be covered by different collective agreements depending on firmcharacteristics; for example, small and large textile firms are subject to different agreements.
8
omnes, meaning that every worker is subject to the provisions of a collective agreement, indepen-
dent of whether she is enrolled in a union.12 Collective agreements are, for all purposes, equivalent
to legislative provisions; hence, in absence of a minimum wages an employer is sanctioned if the
collective agreement in force is violated.
Recent empirical work has shown that collective bargaining is a major determinant of com-
pensation policies, making nominal wages largely invariant across firms or areas.13 For example,
Boeri et al. (2017) note that the gap in nominal wages between North and South Italy is 4% despite
differences in productivity in the order of 30%. They compare the Italian labor market with the
German one, where the increase in pay dispersion has been much more pronounced (Card, Heining,
and Kline (2013)).14.
Importantly, compensation schemes in Italy tend to be “two-tier” (Dell’Aringa (2017)), whereby
wages are given by the sum of a minimum wage, prescribed by the relevant collective agreement,
and a firm-specific top-up. This implies that any change in the minimum wage will affect the entire
distribution of earnings.15 Of course, firms could mitigate or even fully eliminate the effects of
changes in minimum wages by adjusting individual top-ups. However, in Section 4.4 we show em-
pirically that minimum wages established by collective agreements do generate significant variation
in actual workers’ compensation.
4.2 Data
Our dataset combines and matches several sources. We use a unique dataset on the universe
of all the Italian individuals working in the private sector through restricted, on-site access to a
confidential database made available by Istituto Nazionale della Previdenza Sociale (INPS), the
12From Article 39 of the Italian Constitution: Registered trade unions (...) may, through a unified representationthat is proportional to their membership, enter into collective labor agreements that have a mandatory effect for allpersons belonging to the categories referred to in the agreement.
13In addition to surveying the existing literature, we will provide additional corroborating evidence in Section 4.4.14See also Devicienti, Fanfani, and Maida (2016) for worker-level evidence. These nominal rigidities have important
consequences for real wages, causing for example large urban wage premia (Belloc, Naticchioni, and Vittori (2018)).15Boeri (2015) shows notes that such arrangement, fairly common in Western Europe, fails to align wages and firm
productivity.
9
Italian national institute of social security.16 INPS has a number of tasks, such as delivering
retirement or disability pensions, as well as unemployment benefits.
Starting from January 2005, the dataset records income and other information of each worker
at a monthly rate. An individual worker may be employed under multiple contracts, possibly
with the same firm, and therefore may appear in the dataset more than once in a given month.
Crucially for our analysis, the dataset also has a variable called contract code, which identifies the
job category each individual worker’s job belongs to. Since collective agreements are, de facto,
equivalent to legislative provisions in case of a labor lawsuit, such a variable is recorded with a high
degree of accuracy. We retrieve information on each collective agreement from the web archive
of Consiglio Nazionale dell’Economia e del Lavoro (CNEL), a council, established by the Italian
Constitution, that advises the executive and legislative branches of the public administration. For
each of the over 2,000 contracts available in the archive, we code the dates in which the contract
was signed and expired, as well as possible vacancy periods.
Since the workers’ contracts also records the fiscal identifier of the company, we can merge
the INPS data with the Bureau Van Dijk (BvD) “Amadeus” database. We purchase intraday data
on stock prices of all the companies listed on the Milan Stock Exchange between 2005 and 2016
from Tick Data, LLC. Since we rely on stock market reactions over a relatively short window, we
exclude firms with stocks that are very illiquid, namely those for whihc the average price over the
sample is below 1 euro. This overall merging of firm-level data allows us to link the contract-based
variables with information on the firms’ values. We further merge the combined INPS-Amadeus
dataset as well as the stock market data with Compustat Global through ISIN or firm name to
obtain the accounting information of the listed companies.
Finally, we use intraday data on EONIA swap rates.17 We refer to Section 4.5 for details about
how we use swap rates to construct our monetary policy shocks.
16The data was made available to selected researchers through the project “VisitINPS”.17Kindly shared with us by Corsetti et al. (2018).
10
4.3 The Wage Rigidity Proxy
One contribution of our paper consists in the precise identification of the nominal rigidities through
a fully exogenous proxy of wage rigidity, which varies across firms. To obtain this we aggregate, at
the firm-month level, information on workers’ contract termination dates, occurring prior to their
contract renewal.
For any given job category c and day t, we know the date on which the agreement in force
was signed, its date of expiration τt and the date on which the agreement that replaced the one in
force on date t was signed. However, the latter and the date of expiration often do not coincide
because of “vacancy periods” during which, as long as employers’ representatives and trade unions
cannot find an agreement and sign a new contract, the expired contract is assumed to be in force.
We could, in principle, construct the variable “time to renewal” of each agreement simply by
subtracting t from the date on which the contract that replaces the one currently in force is signed.
This would introduce, however, a forward-looking bias because, on day t, agents are not aware of the
date of the next renewal - i.e., they do not know if there are going to be any delays in the bargaining
process. They foresee, however, when the agreement is going to expire. Thus, we adjust our “time
to renewal” variable by subtracting the current date from the expiration date. We truncate this
variable to zero to avoid negative values which would not be meaningful for economic purposes and
which would otherwise arise during vacancy periods. Intuitively, we realistically conjecture that
investors form their expectations assuming that, after the expiration of an agreement, the next
agreement will be signed any time soon.
Any firm can hire workers in several job categories, who are therefore subject to different
collective agreements. Based on this, we construct a firm level measure of wage rigidity WR as
follows:
11
WRi,t ≡ log
1 +
∑j
wi,j,c ×max {0, t− τi,j,c,t}∑j
wi,j,c
(3)
where j, t, c index workers, firms and job categories, respectively. We use the max operator to
truncate the difference t− τt at 0 for each worker, and take its average at the firm level using the
wage earned in the previous month as weight18. Because of the truncation at zero, the measure
is right-skewed. We re-normalize the distribution by adding the wage rigidity-measure to the
logarithm of 1.
It would not be feasible to construct a wage rigidity metric with high degree of accuracy without
the use of administrative data. In most cases indeed there is no mapping between standard industry
definitions and sectors, as defined by collective agreements. More importantly, as Figure 2, Panel
A shows, the vast majority of firms employs workers subject to at least two collective agreements.
Panel B shows, instead, the fraction of total wages paid to workers subject to the modal contract;
the median firm has a sizeable fraction, about 20%, subject to a contract other than the most
prevalent. Hence, once more, an accurate aggregation of wages would not be possible without
observing the exact job categories.
A potential concern could be that vacation periods could be not orthogonal to factors that
could, in turn, be priced on days of monetary policy announcements. For example, such renegotia-
tions may become more difficult and time-consuming during times of industry downturns. However,
during economic slowdown one would expect the sensitivity of stock prices to macroeconomic shocks
to be higher, because firms with low profitability are likely to be less able to absorb temporary
changes in cash flows. In our empirical results, we find that the sensitivity drops when WR is
low, and possibly zero. More importantly, in Appendix A we test directly whether firms’ perfor-
mance can systematically predict vacancy periods. As Table A1 shows, neither sale growth nor
18This choice is preferable to that of assigning an equal weight to each worker, as it better reflects the actual laborcost of each individual.
12
operating performance are associated with higher or lower likelihood of observing vacancy periods,
attenuating concerns of omitted variable bias.
4.4 Minimum Wages and Workers’ Compensation
Our wage rigidity proxy can account for nominal rigidity only if actual wages are closely linked to
minimum wages established by collective agreements. This connection should be implied by the
“two-tier” structure of individual compensation contracts. Nevertheless, our data allow us to test
this link formally.
We manually code, for each contract, the minimum wage at any given month between 2005
and 2016.19 We then regress the logarithm of worker’s monthly compensation on the logarithm of
the minimum wage relevant for her collective agreement, year-month and worker-firm dummies.20.
We include workers of listed firms21. Overall, we have 37,154,864 worker-month observations and
estimate the following model:
log(wi,t,c) = β log(wMc,t) + γi + δt + ωc + εi,c,t (4)
where i, c and t identify worker, job category and year-month, respectively. w is the worker’s
wage, wM the minimum wage established by the relevant collective agreement, and γ, δ and ω
are dummies for worker, year-month, and job category, respectively. The estimates of β is large,
equal to 0.763, and highly statistically coefficient, being the standard error equal to 0.120 and the
R2 = 0.715.22 Hence, the pass-through from minimum wage to actual compensation is substantial,
even for workers who may earn substantially more than the minimum.
A last, we wish to address the possible reverse causality between actual wages and minimum
wages. High growth in an industry may push wages and, at the same time, allow unions to obtain
19We thank INPS and, in particular, Maria Cozzolino, for helping us in this fairly time-consuming task.20Collective agreements establish minimum wages for different job titles, which are ordered in “level” depending
of the position in the firm hierarchy. Job titles are not however recorded by INPS; hence, we take the “minimum ofthe minimum wages” for each collective agreement.
21We exclude part-time workers since we do not observe hourly compensations.22See also Fanfani (2019), who studies a different sample and employs a different empirical specification but obtains
qualitatively similar results.
13
concessions from industry representatives in terms of higher minimum wages. Importantly, prof-
itability in a sector is likely to follow long-term trends, and so to evolve gradually. On the other
hand, minimum wages move in steps; that is, they ”jump” in given months, according to a prede-
termined schedule. Hence, if changes in minimum wages are not led by changes in actual wages,
we could be less concerned about a potential spurious correlation. We examine the relationship
between minimum wages and actual compensation by estimating a dynamic model along the lines
of Dube, Lester, and Reich (2010):
log(wi,t,c) = β16∑
j=−8
(η−6j∆6 log(wMc,t+6j)
)+ η−96 log(wMc,t−96) + γi + δt + ωc + εi,c,t (5)
In this equation, ∆6 represents a a six-month difference operator23 Figure 3 plots the resulting η
coefficients and, hence the cumulative effect of minimum wage changes on workers’ compensation.
All the coefficients on the leads are close to zero and statistically insignificant, but there is a “jump”
at time t, and the effect does not vanish in the long run. To summarize, collective agreements are
a significant source of variation in individual wages and, hence, are likely to be priced in the stock
market response to monetary policy shocks.
4.5 Monetary Policy Shocks
We follow recent high-frequency identification methodologies for monetary policy shocks. The ra-
tionale behind it is that because markets only price new information, it is possible to extract the
unexpected component of monetary policy decisions by measuring high frequency movements of
market prices in a narrow window of the policy announcements. We obtain the list of announce-
ments of the ECB target rates from the ECB website and match these dates with the corresponding
changes in the 1-year Euro Overnight Index Average (EONIA) swap rate, our proxy for monetary
policy shocks. The 1-year rate strikes the best balance between being highly sensitive to monetary
policy announcements, while remaining relatively unaffected by term premia24.
23We have 48 months of leads and 96 of lags. This asymmetry arises since our sample ends in 2016 and we havedata on collective agreements only until 2018. Shortening or extending this window produces similar results.
24In Section 5.2 we show that our results are not sensitive to this particular choice.
14
An important choice is related to the window around the monetary policy announcement.
The ECB target rates are announced at 13:45 CET. However, unlike in the US, investors learn
much about the future course of ECB actions during the press conference, followed by a Q&A with
the president, which starts 45 minutes later, at 14:30 CET. Hence, an appropriate window should
surround both the actual announcement and the press conference.
We follow Corsetti et al. (2018), who choose a 6-hour window from 13:00 to 19:00 CET.
This matches the closing times of the Tokyo and London stock exchanges, respectively. While a
shorter window could purge stock returns from noise unrelated to monetary policy announcements,
some of our stocks are relatively illiquid and infrequently traded, and hence may not react to new
information instantaneously, unlike the components of the S&P 500 studied by Gorodnichenko and
Weber (2016), or sovereign bold yields and major stock market indexes, as in Altavilla, Brugnolini,
Grkaynak, Motto, and Ragusa (2019). Hence, our choise is a compromise between the narrower
windows used in these studies and the wider (daily) ones used in other settings (for example in
Bernanke and Kuttner (2005)). We will return on this issue in Section 5.2, where we reestimate
our baseline regressions using the dataset of high-frequency monetary policy shocks recently made
available by Altavilla et al. (2019).
Given the specifications above we measure the monetary policy shock as follows:
MPt ≡ S+t − S
−t (6)
where the index t in S indicates the date of an ECB meeting, when ECB target rates are announced,
and S+t and S−t are the 1-year Euro Overnight Index Average (EONIA) swap rates, shortly after
and before the time of the announcement, respectively.
5 Empirical Evidence
5.1 Data and Econometric Strategy
We wish to test the two main implications of the simple model derived in Section 3. First, we
test whether the impact of a monetary policy shock is larger when the distance from contract
15
renegotiation, hence wage rigidity, is higher. Second, to assess the distributional consequences
of monetary policy, we examine how the impact varies depending on firms’ labor intensity. Our
baseline econometric specification reads as follows:
R2i,t = αMP 2
t + βWRi,t + γMP 2t ×WRi,t + δ′Xi,t + θ′Xi,t ×MP 2
t + ηi + µt + εi,t (7)
where WR is and MP , that proxy for wage ridigity and monetary policy shock, are defined in
equations 3 and 6, respectively. The variable R2i,t is the square of the firm’s stock return on an
announcement day, defined as the price of the stock at 19:00 CET minus the price at 13:00 CET, all
divided by the latter. As explained in Section 3, we square this variable as we do not have enough
information on the direction of the change in firms’ profits (upward or downward). Hence, the best
and the safest is to measure the effects of monetary policy shocks on the volatility of firms’ returns.
The monetary shock MP 2 is the squared difference between the swap rate on the target policy rate
at 19:00 CET and 13:00 CET. At last, we include an interaction term between the monetary policy
shocks and the wage rigidity measure. Our key prediction is related to the sign of the coefficient
γ, which we expect to be positive.
The econometric specification also includes a vector X of control variables that could be
potentially related to firm volatility on announcement days. Those are as follows. Size is the
logarithm of total assets. Leverage is defined as the sum of debt in current liabilities plus long-
term debt, all divided by total assets. ROA is earnings before interest, debt and amortization
divided by total assets. All these variables are measured at the beginning of the fiscal year in which
the announcement is made. Labor Intensity is defined as the total wage expenses in the month
preceding the announcement, divided by total assets at the beginning of the fiscal year. Our final
sample has 25,529 observations and comprises all the monetary policy announcements occurred
between 2005 and 2016. Descriptive statistics for the main variable used in the paper are in Table
2.
Importantly, we also include the same vector interacted with the monetary policy shock, to
16
account for the possibility that our proxy WR is co-moving with some other firm characteristics.
Finally, in most of the regressions we also include the vectors of firm and day dummies ηi and µt,
respectively. The time-fixed effects, whenever included, will absorb the coefficient on MP 2 but not
our main coefficient of interest.
5.2 Results
Estimation results are shown in Table 3. We estimate six variations over the baseline equation,
going from the least to the most conservative specification. For ease of interpretation, we divide
all variables, before taking interactions, by their sample standard deviations after subtracting their
means. In column 1 we include only WR and MP 2, as well as the interaction between the two,
which is our coefficient of interest. Consistent with Gorodnichenko and Weber (2016), the coefficient
on MP 2 has a positive sign, and is significant at the 1% level. The standalone coefficient on WR,
that does not have an obvious economic interpretation, is instead insignificant.
Crucially, the coefficient on the interaction term MP 2 ×WR is positive, and equal to 0.017,
significant at the 1% level. The positive sign on this term is in line with the analytics of the simple
model described above, as well as with the extended dynamic model that we will present in Section
6. Given that the coefficient on MP 2 is equal to 0.040, the two coefficients are in the same order of
magnitude. In column 2 we add as controls size, leverage ROA and labor intensity. We find that
firms with higher debt levels and lower profitability exhibit higher volatility during announcement
dates. The coefficient of interest is essentially unaffected.
In column 3, we interact MP 2 with the control variables. This allows us to account for the
possibility that WR is simply proxing for, or strongly correlated with, some firm’s characteristics.
Our coefficient of interest is slightly reduced in magnitude but remains significant at the 10% level.
Hence, the influence of our measure of wage rigidity on firms’ sensitivity to monetary policy shocks
appear largely orthogonal to that of the most plausible firms’ characteristics.
In columns 4 through 6 we add to the specifications of columns 1 through 3 both firm and
17
announcement date dummies. Hence, the coefficient on MP 2 is not separately identifiable, but
its interaction with the wage rigidity proxy still is. The inclusion of firm fixed effects allows
us to control for time-invariant firm characteristics that may determine stock volatility during
announcement dates. The inclusion of those two vectors of fixed effects only slightly increases the
size of the coefficient of interest, which now varies between 0.021 and 0.022 and remains significant
at the 1% level. Among the control variables, the only one that remains significant, and negative,
is the interaction term MP 2 × Size.
In Tables 4 and 5, we propose several robustness tests. For brevity, we report only estimates
from our most conservative specification, which includes firm and date-fixed effects, as well as the
control variables, both as standalone regressors and interacted with the monetary policy shock.
First, we show that our results are unchanged when we measure monetary policy shocks using
changes in swap rates at different horizons. Columns 1 through 3 of Table 4 show that results
are essentially unaffected independently of whether we use changes in 1-year, 6-month or 3-month
swap rates.
Although workers in a given firm may be subject to different collective agreements, in practice
they tend to cluster across industries. Hence, one concern may be that our results might reflect
an “industry effect”. An attractive feature of our econometric strategy is that within industry
our stickiness measure changes over time. Thus, in our second robustness check we control non-
parametrically for a possible industry effect by interacting the monetary policy shock with industry
dummies. Industry are defined using the 1, 2 or 3-digit NACE industry classification, which is
standard in the European Union. As columns 4 through 6 of Table 4 show, the coefficient of interest
remains similar in magnitude, and significant at the 1% level. Hence, we can safely conclude that
our measure is immune from this concern.
Finally, we test whether our results depend on our particular window choice for the monetary
policy shock and the stock return. In Table 5 we make use of the dataset made available by
Altavilla et al. (2019). Our monetary policy shock is now the squared change in the median quote
18
of the EONIA swap rate from the window 13:25-13:35 (before the press release) to the median
quote in the window 15:40-15:50 (after the press conference). In column 1, we measure the stock
market response using the same window. Results are qualitatively similar, although the magnitude
is slightly lower (0.012, significant at the 5% level). In column 2 we retain the advantage of this
narrower window for the measurement of the monetary policy shock, but go back to the 6-hour
window for the dependent variabile. The coefficient is now 0.017, significant at the 1% level, and
very similar to the one estimated in the baseline regressions. Finally, in column 3 we use daily
stock returns; again, the coefficient is similar and remains significant (0.015, signigicant at the 5%
level). To summarize, results do not appear to be dependent on the particular window choice for
the measurement of the monetary policy shock; however, a too narrow window for the stock returns
tends to reduce the precision of the estimates. This is precisely what we would expect in presence
of stocks that do not necessarily trade very frequently and, hence, may not react instantaneously
to new information.
5.3 The Labor Channel
To further validate the economic channel we posited, we analyze the cross-sectional heterogeneity
of our results. Our simple model of Section 3 highlights that the strength of the wage rigidity
channel is larger if the workforce extracts a larger share of the surplus., i.e., if the β of our model
is large.
We test this prediction by sorting firms according to their degree of labor intensity, measured
in two different ways. First, as in Table 3, we use the ratio of wage expenses divided by total assets.
Second, we use the total number of days worked, again divided by total assets. Hence, in the first
case the labor intensity proxy is based on actual labor costs, while in the second case it is based on
the size of workforce employed. As usual, both numerators are measured in the month predating
the announcement, and are scaled by total assets at the beginning of the fiscal year.25
25Total sales would perhaps be more coherent with the model presented in Section 3 as denominator of the proxyfor labor intensity. We chose total assets because the variable sales is missing in about 20% of the observations. We
19
Table 6 shows regression coefficients obtained after estimating equation 7 in the subsamples of
firms characterized by labor intensity in the bottom or top 50%, respectively. Results are consistent
with our prediction. In column 1, the coefficient on MP 2 ×WR is small (0.009) and insignificant.
On the other hand, in column 2, namely the case of high labor intensity firms, it is large (0.025)
and significant at the 1% level. Columns 3 and 4 display similar results, with the coefficient of
interest rising from an insignificant 0.009 to 0.023, significant at the 5% level. This test represents
an important validation of our hypothesis, as it shows that the labor demand channel is the crucial
one in our narrative.
As a “placebo” test, we replicate our main tests by using, as dependent variables, the squared
stock returns on the day before an actual policy announcement. We replicate both the base-
line regression presented in Table 3 and the heterogeneity analysis presented in this section. As
Appendix-Table A2 shows, all the coefficients are insignificant and no apparent relationship between
wage rigidity and labor intensity emerges.
5.4 Firm Profitability, Business Cycle and Flexible Staffing Arrangements
In Table 7, we verify whether firms’ conditions or the business cycle affect the importance of wage
stickiness in determining the stock market response to monetary policy shocks. It is reasonable
to expect that firms whose profits are lower will be less able to shelter the impact of the shock.
Similarly, during economic downturns, it is foreseeable that rigidities in labour costs would hamper
firms’ ability to adjust labor costs. Hence, their profits would be more sensitive to the monetary
policy surprise.
In columns 1 and 2 we use ROA as a sorting variable, to distinguish between highly and non-
highly profitable firms. The wage rigidity channel appears stronger for firms with low profitability.
The coefficient for this group of firms is highly significant and equal to 0.028. The coefficient drops
to 0.010 and is insignificant in highly profitable firms, instead. In columns 3 and 4, we use an
have, in any event, re-run our analysis using labor expenses over revenues as sorting variable and, in this smallersample, found very similar results.
20
alternative sorting variable, namely the quarterly GDP growth. Again, the results appear to be
driven by shocks occurring at times of (relative) economic distress. In column 3 the coefficient is
0.020, significant at the 5% level, while in column 4 the coefficient drops to -0.03 and is insignificant.
In Table 8, we compare firms according to the ease in which they can adjust the workforce on
the extensive margin. A firm constrained to a fixed wage schedule could more easily absorb the
impact of a demand shock when its workforce is subject to more flexible contractual arrangements
(Houseman (2001)). We test this hypothesis by exploiting information on workers’ contract types.
Friesen (1997) shows that part-time, as opposed to full-time labor is a source of flexibility when
facing economic shocks. Similarly, fixed-term contract also provide an important margin of flexibil-
ity to a firm. A firm may simply let expire part of the current fixed-term contracts when facing a
negative shock, hence being able to easily reduce the workforce. Being characterized by low firing
costs, such contracts also allow a firm to, simmetrically, expand its workforce following a positive
shock with little long-term commitment towards the new hires.
Our dataset has detailed information on the type of contract each worker is subject to. Hence,
we can construct measures of flexibility at the firm level, depending on the fraction of workers who
have a part-time contract or a fixed-term contract. In 8, we sort firms according to eiher measure.
Column 1 shows that in firm with a fraction of part-time workers below the median, the coefficient
of interest is 0.021, significant at the 1% level, but drops to an insignificant 0.05 in firms that are
above the median. Results are qualitatively similar, although less stark, when looking at firms that
differ in their usage of fixed-term contracts. In this case the coefficient on the interaction term
of the wage rigity measure with the monetary polcy shock moves to a significant 0.024 to 0.017,
significant at the 10% level.
The robustness tests described above also strengthen the relevance of the distributional conse-
quences of monetary policy shocks. Changes in the interest rate more strongly affect firms that face
additional rigidities in labour cost adjustments. Our results complement also other recent evidence
on the effects of monetary policy with heterogenous firms. Bahaj et al. (2017), using a panel of UK
21
firms, examine the impact of monetary policy on employment dynamics for small and large firms,
pointing at the role of different financial frictions.26 Other works (see, for example, Rodnyansky
(2018)) examine the differential impact of monetary policy for exporting and non-exporting firms.
5.5 Real Effects
Since our focus is on an employment channel, it is paramount that we also test the sensitivity
of labour demand to the monetary policy shock. This would directly validate our arguments.
Therefore, we estimate a new specification of the model whereby the dependent variables captures
various aspects of the labour demand, including intensive and extensive margins. Our dependent
variables now are the squared symmetric growth rates of four different objects. “Pay” is the total
compensation, given by the sum of all wages paid to the firm’s employees. “Days worked” is defined
as the total of workers employed times days worked. This captures changes in the intensive margins.
“Employees” is the total number of workers (i.e., “bodies”) employed. This captures the extensive
margin. Finally, “full-time employees” are those who are employed for at least 20 days per month.
To run our new specification we shall harmonize the timing of observation of the variables.
Hence, we first aggregate our monetary policy shocks at the quarterly level, following the procedure
recommended by Gertler and Karadi (2015). We first define et as:
et ≡0∑−92
et (8)
and then cumulate the ets over each quarter Q to obtain a quarterly shock eQ:
eQ =∑t∈Q
et (9)
Intuitively, this procedure assigns higher weight to shocks that occur early in the quarter, and that
therefore have more time to display their effects.
At last, note that we do not have plant-level information of the firms’ workforce. This prevents
us from identifying and excluding changes in employment due to divestitures or acquisitions. This
26Past works along the same lines include Gertler and Gilchrist (1994), Moscarini and Postel-Vinay (2012), Kudlyakand Sanchez Kudlyak and Sanchez (2017) and Mehrotra and Crouzet (2017), among others.
22
means that our outcome variables are prone to exhibit outliers. As a partial remedy, we winsorize
the dependent variables at the 5% level.
Results for those new specifications are displayed in Table 9. As usual, all the variables are
demeaned and divided by their standard deviation. We find significant and economically large
effects on each of the first three outcome variables. In particular, we find that the coefficients of
interest are significant at the 5% or 1% level, with estimates between 0.029 and 0.037. Interestingly,
the coefficient of interest becomes insignificant only for full-time employees. Its magnitude, 0.015, is
much smaller than the one found in column 3 (equal to 0.037), where the change in total employment
is used as dependent variable. This adds an interesting dimensions and suggests that managers
respond to aggregate shocks by primarily adjusting the number of short-term employees.
We find significant and economically large effects on each of the first three outcome variables.
In particular, we find that the coefficients of interest are significant at the 5% or 1% level, with
estimates between 0.029 and 0.037. Interestingly, only in the last column, where we focus on long-
term employees, we find an insignificant coefficient. Its magnitude, 0.015, is much smaller than the
one found in column 3 (equal to 0.037), where the change in total employment is used as dependent
variable. This suggests that managers respond to aggregate shocks by primarily adjusting the
number of part-time employees.
We also run a “placebo test” along the lines of the one described in Section 5.3. Specifically
we replace the dependent variables with the their one quarter-lags equivalent. As Appendix-Table
A3 shows, the coefficients are now small and insignificant.
6 A Multi-Sector Model with Sectorial Wage Rigidity
Our empirical analysis was guided by the predictions of a simple model. Our estimates have
also highlighted further twists of the employment channel. Equipped with these we can now
construct a more sophisticated micro-founded dynamic general equilibrium model, which can better
account for general equilibrium spillovers and give better ground to the cross-sectional and temporal
23
dimensions. A structural estimation also allows us to further validate our empirical results through
the lenses of the model. In fact we will use the model-based simulated data and run regressions
whose specification mirrors the ones in the data. The model delivers the same results obtained
with the data regressions.
The model can be generally described as follows. In the model there are households/workers
and firms which are heterogenous in that they confront different degrees of wage rigidity. We start
by describing the production sector. This is composed of a discrete number of sectors, indexed by
s = {1, ..., S}. In each sector there is a continuum of firms producing different varieties, indexed by
i. Firms are monopolistic. Hence firms’ stock values is the sum of future discounted mark-ups. Ex
ante, each firm i demands differentiated labor services from a placement agency. The differentiated
labor services in each sector are then bundled into a composite labor input by the competitive
representative placement agency. In each sector, wages are set by a labor union in staggered fashion.
This implies that firms operating in different sectors feature different marginal cost dynamics. More
specifically, the sectorial labor input is supplied by the households to the sectorial labor union, which
takes its members’ utility into account and acts as a monopoly supplier27 of differentiated labor
services to all the variety-producing firms in the sector. We introduce wage rigidity through a
staggered structure a la Taylor (1979), such that each union can re-set wages only at specific time
intervals, which are different across sectors. Heterogeneous wage stickiness implies heterogeneity
in firms’ marginal cost dynamics, and hence different responses to any monetary policy shocks.
The comprises also a continuum of households which consume, invest in Arrow-Debreu secu-
rities and supply labor services to different sectors and to different variety-producing firms within
each sector. Hence total labor supply comes from aggregating firm level labor services, upon which
unions have monopolistic power, and then sectorial labor services28.
27The assumption through which labor unions hold the monopoly power and hence set wages is realistic and hasbeen advanced before by Schmitt-Grohe and Uribe (2005).
28Even though individual household members supply labor services to different firms and sectors, heterogeneitydoes not affect their consumption-saving decisions. Each household consists of a large number of individuals, eachindividual supplies one unit of labor inelastically and shares all income with the other household members throughinsurance scheme. Arrow-Debreu securities also shelter from idiosyncratic shocks. This specification is along the lines
24
6.1 Households
The household has a lifetime utility function:
Wt = Et
{ ∞∑k=0
βkU
(Ct+k,
S∑s=1
(fs)
∫ 1
0Nsi,t+kdi
)}, (10)
where β is the discount factor, Et is the conditional expectation operator with respect to information
at time t, Ct is a consumption aggregate and Nt denotes labor services. The sum of the households
in the economy supplies labor services to all sectors,∑S
s=1(fs)Ns,t+k, where Ns,t+k =
∫ 10 Nsi,t+kdi.
In each sector, unions will only represent the portion of households whose labor services are supplied
there. Labor supply in each sector is then an aggregate of labor services to the continuum of firms
producing varieties, i. We specify the per-period utility function as (Ct−hCt−1)1−σ
1−σ − ψ(Ns,t)1+η
1+η ,
where σ the elasticity of inter-temporal substitution, η−1 is the Frisch elasticity of labor supply,
and h is an internal habit parameter. We include external habits in the utility function as this
is known to capture well the hump shaped response of output and other variables. However, its
absence does not affect the employment channel we seek to confront with the data.
The representative household maximizes its utility subject to the budget constraint:
CtPt +∑ht+1
νt+1,tBt+1 =S∑s=1
(fs)Ws,t
∫ 1
0Nsi,t+kdi+Bt − τt +
∫ 1
0Γt(i)di, (11)
where the household earns income from differentiated labor Nsi,t at the nominal wage rate Wsi,t
through the labor services supplied by the union. The sectorial union will eventually choose one
wage for all firms i operating in sector s. The final good can be used for either saving in the
Arrow-Debreu security or consumption, and sells at the nominal price Pt. Households also face a
lump-sum tax τ . To insure their consumption pattern against random shocks at time t households
spend νt+1,t Bt+1 in nominal state contingent securities, where νt,t+1 ≡ ν(ht+1|ht) is the period
t price of a claim to one unit of domestic currency in state ht+1, divided by the probability of
occurrence of that state. Each asset in the portfolio Bt+1 pays one unit of domestic currency at
of Schmitt-Grohe and Uribe (2005) and Galı (2015).
25
time t + 1 and in state ht+1. Finally, Γt(i) is the profit of monopolistic firm i, whose shares are
owned by households.
The representative household chooses processes {Ct}∞t=0 and bonds {Bt+1}∞t=0, taking as given
the set of processes {Pt, νt+1,t}∞t=0 and the initial wealth B0 so as to maximize (10) subject to (11).
For any given state of the world, the following set of efficiency conditions must hold:
λt = (Ct − hCt−1)−σ − βh(Ct+1 − hCt)−σ (12)
βPtPt+1
λt+1
λt= νt,t+1 (13)
limk→∞
Et {νt,t+k Bt+k} = 0 (14)
where λt is the Lagrange multiplier on the budget constraint, equation (13) is the Euler condition
on the state-contingent bond holding for each state of the world, and equation (14) is a condition
on terminal wealth.
Note that labor decisions are made by a central authority within the household, a union, which
supplies labor monopolistically to a continuum of labor markets of measure 1 indexed by j ∈ [0, 1].
The derivations of the optimal supply of each labour service, given the demand by firms and the
unions’ optimization problem is derived in sections 6.2.3 and 6.2.4. This set-up of imperfectly
competitive labor markets follows Schmitt-Grohe and Uribe (2005) and Galı (2015) and retains the
advantage of avoiding that the heterogeneity across households in the number of hours worked spills
over into consumption heterogeneity. This characterization allows us to retain firms’ heterogeneity
in labour demand, the salient channel through which we wish to characterize the heterogenous
effects of monetary policy, without adding complexities to the consumption profiles.
6.2 The Final Good Sector
Households consume a homogeneous good, which results from bundling, through a Dixit and Stiglitz
(1977) aggregator, the continuum of varieties produced by firms, indexed by i, in each sector. In
each sector, a representative competitive firm bundles together a Dixit-Stiglitz composite of varieties
26
produced by monopolistically competitive firms. At the aggregate level a representative competitive
firm bundles together a Dixit-Stiglitz composite of the sectorial outputs. The optimization problems
solved by each of those firm-level and sectorial-level output bundlers will deliver the optimal demand
for each variety i, which in turn will depend upon the optimal demand of the sectorial product s.
The optimal demand for each sectorial variety will represent the constraint of the monopolistically
competitive firm i, choosing prices in sector s. Note for sake of clarity that monopolistic firms
operating in each sector will have to be indexed both by the sectorial index, s, and by their
individual variety index i ∈ [0, 1]. Each sector size is given by fs > 0, with∑S
s=1 fs = 1.
Given the above, the sectorial output is given by Ys,t =
[(fs)
εp−1
εp∫ 10 Y
εp−1
εp
si,t di
] εpεp−1
, where Ys,i,t
is the variety produced by firm i in sector s. The parameter εp ≥ 0 is the elasticity of substitution
within as well as across sectors. The aggregate output is given by Yt =
[∑Ss=1(f
s)1εp Y
εp−1
εp
s,t
] εpεp−1
.
The output-bundlers at sectorial level solve an optimization problem29 to obtain the downward-
sloping demand for sectorial varieties s = {1, ..., S}, which reads as: Ys,t = fs(Ps,tPt
)−εpYt. Sim-
ilarly, the representative firm that bundles individual varieties into sectorial goods solves an op-
timization problem30 to obtain the downward-sloping demand for individual varieties i ∈ [0, 1],
which reads as: Ysi,t = (f s)−1(Psi,tPs,t
)−εpYs,t. Given the optimal demands for varieties and sectorial
output, we can derive the price index Pt =[∑S
s=1(fs)P
1−εps,t
] 11−εp , and the sectorial price aggregator
is Ps,t =[∫ 1
0 P1−εpsi,t di
] 11−εp .
29The problem is rather standard and reads as follow:
maxYt
[PtYt −
S∑s=1
Ps,tYs,t
]
subject to Yt =
[∑Ss=1(fs)
1εp Y
εp−1
εps,t
] εpεp−1
.
30The optimization problem reads as follows:
maxYsi,t
[Ps,tYs,t − fs
∫ 1
0
Psi,tYsi,tdi
],
subject to Ys,t =
[(fs)
εp−1
εp∫ 1
0Y
εp−1
εp
si,t di
] εpεp−1
.
27
6.2.1 Intermediate Goods in each Sector
Each variety Ys,i,t is produced by a single firm in a monopolistically competitive environment.
Each firm i ∈ [0, 1] produces output using labor services Ns,i,t as factor inputs. The production
technology is Ysi,t = Nsi,t.
For robustness reasons, we foresee two alternative specifications for labour demand. In the
first firms simply equate marginal returns to labour to marginal costs. In the second, we introduce
convex hiring costs. In this second case firms choose labour demand by minimizing discounted
future labor costs Ws,tNs,t+ξ2Ns,t
[Ns,tNs,t−1
− 1]2
, subject to Ysi,t = Nsi,t. The demand for individual
labour varieties in this case yields:
mcs,t =
[Ws,t +
ξ
2
(Ns,t
Ns,t−1− 1
)2
+
(ξNs,t
Ns,t−1− 1
)− βξEt
λt+1
λt
(Ns,t+1
Ns,t− 1
)N2s,t+1
N2s,t
]/Pt, (15)
Here the real marginal cost mcs,t, namely the Lagrange multiplier deflated by the price level, is
given by the sector’s real wage plus an adjustment cost term that reflects hiring costs.
Wages in turn are determined in a staggered fashion in the spirit of Taylor (1979). This is
derived and discussed in section 6.2.4.
At last, monopolistic firms choose prices to maximize the stream of real profits, Πsi,t =
Psi,tPtYsi,t − Ws,t
PtNsi,t, subject to (15), Ysi,t = Nsi,t and the demand for individual varieties. The
optimality condition reads as follows:
Psi,t =εp
εp − 1mcsi,tPt, (16)
Prices are set equal to a mark-up on nominal marginal costs. In a symmetric equilibrium we can
impose Psi,t = Ps,t, Nsi,t = Ns,t and mcsi,t = mcs,t.
6.2.2 Firm’s Value
As one of our goals is to assess the effects of monetary policy shocks on firms’ values, we derive
the model-equivalent here. Imposing symmetry across firms i within sector s, we derive a recursive
28
formulation for firm’s values in sector s as follows:
Vs,t = Πs,t + βEt{λt+1
λtVs,t+1
}. (17)
Sectorial returns are the percentage change in the sectorial value:
Rs,t = log(Vs,t)− log(Vs,t−1) (18)
6.2.3 Labor Services Aggregator
Wages are set by monopolistic unions, each operating within the sectors. Unions have monopoly
power over the individual labor services that household members supply to the continuum of firms.
Unions choose wages subject to the demand for each labor service, i, which we derive next.
In each sector a competitive labor agent bundles together the labor services demanded by each
firm i. The Dixit-Stiglitz aggregator for the sectorial labor demand isNs,t =[(fs)
εw−1εw
∫ 10 (Nsi,t)
1−εwεw di
] εwεw−1
,
where εw is the elasticity of substitution between labor varieties within sectors. Similarly, at
the aggregate level a competitive labor-placement agent bundles together the labor services sup-
plied at the sectorial level. The Dixit-Stiglitz aggregator for economy-wide labor demand is
Nt =
[∑Ss=1(f
s)1εwN
εwεws,t
] εwεw−1
, where εw is also the elasticity of substitution between labor va-
rieties across sectors. The competitive labor-placement agent in each sector solves an optimization
problem31 yielding the downward-sloping demand curve for labor variety i in sector s, namely
Nsi,t = (fs)−1(Wsi,t
Ws,t
)−εwNs,t. Similarly, the competitive placement agent at the aggregate level
chooses the optimal labor demand32, which results in Ns,t = f s(Ws,t
Wt
)−εwNt.
31It chooses the optimal demand, Nis,t, to maximize:
Ws,tNs,t −S∑s=1
Wsi,tNsi,t
subject to Ns,t =[(fs)
εw−1εw
∫ 1
0(Nsi,t)
1−εwεw di
] εwεw−1
.32The placement agent maximizes WtNt − Ws,tNs,t subject to the demand constraint Nt =[∑Ss=1(fs)
1εw N
εwεws,t
] εwεw−1
.
29
6.2.4 Labor Unions and Wage Stickiness
Wage setting is staggered in the spirit of Taylor (1979). In each sector labor unions set the wage
at a specific date and keep it fixed for S periods. In each sector the wage setting date is different.
This creates the staggered structure. This also implies that in each period t only a fraction 1/S of
all households sees its wage adjusted.
Unions choose the optimal wage W ∗s,t by maximizing household utility (10), subject to the
relevant part of the budget constraint (11) and the downward-sloping demand curve for labor
variety i in sector s:
W ∗s,t =εw
εw − 1
Et{∑t+S−1
τ=t βτ−tUNs,τWεwτ Nτ
}Et{∑t+S−1
τ=t βτ−tλt/PτWεwτ Nτ
} , (19)
where UNs is the marginal dis-utility of labor in sector s. Households’ optimal wage is given by a
markup, εwεw−1 , over the ratio of weighted marginal utilities of leisure to marginal utilities of income
within the duration of wage contracts, with the weights given by the normalized demand for its
labor services. If the household expects an increase in the marginal utility of leisure or a fall in the
marginal utility of income within the next S periods, the union will respond by setting a higher
nominal wage.
6.2.5 Firm Wage Heterogeneity
Given that wages are set at different dates in each sector and are kept constant for a fixed interval
S, firms operating in each sector experience different marginal costs. They will thereby react
differently to shocks.
In each sector wages are set according to the general functional form given by (19) but,
operationally, we stagger the sectorial wages in time. Specifically, the wage for sector s = 1 is set
in t = τ , the wage for sector s = 2 in t = τ − 1, ..., the wage for sector s = S in t = τ − S + 1.
Hence, in each period t a firm in sector s pays nominal wages W ∗s that have been set either in the
current or in one of the past periods. This way of staggering the wage contracts over time allows
30
us to tie firms’ marginal costs, hence profits, to the timing in which the monetary policy shock hits
the economy.
To fix ideas, consider a monetary policy shock that occurs at the current period. Firms
belonging to a sector whose renegotiation took place in the previous period are able to adjust their
wage bill only S−1 periods from now. Hence their marginal costs are tied to nominal wages W ∗s,t−1
until then. On the other extreme, firms belonging to a sector whose renegotiation takes place in
the current period will experience an immediate change in their marginal costs, and hence in their
profits. These firms are affected by wage stickiness only through general equilibrium effects, which
they take into account when setting their prices.
6.2.6 Market clearing, Monetary Policy and Calibration
Equilibrium in the model is completed by a market clearing condition, Yt = Ct, and an operational
rule for monetary policy, which reads as follows: 1 + it = (1−ρi)(1 + i) +ρi(1 + it−1 + bW (Wt/W −
1)) + εit, wherein W represents the steady-state value of the nominal wage index.
We solve the model numerically. Out goal is to use simulated data to estimate a model-
counterpart of the regressions estimated in the empirical part. We calibrate the model to quarterly
data using standard values in the literature, given in Table 10. Notably, we choose S = 8 sectors
in order to roughly match the approximate time span of wage contracts in the data. A value of
S = 8 implies that the median renegotiation time is 3.5 quarters, i.e. roughly 315 days, which is
very close to the median of 318 days in the data. In the specifications with hiring costs, ξ is set
to 2.015, which is the average of the values estimated in Cooper and Willis (2009) for positive and
negative employment changes.
6.3 Results
We simulate the model in response to a monetary policy shock active for 2,000 periods. With 1,200
periods used as burn-in, the remaining T = 800 periods are used to obtain artificial data and to
31
sample across sectorial firms’ returns so as to run the following regression:
R2s,t = αεi2t + βWRs,t + γεi2t ×WRs,t + us,t, (20)
Note that for the model-based regression we do not need to include controls, as the simulated data
results from the solution of a model that accounts for general equilibrium effects. The variables
R2s,t represents again the squared returns of the firms producing intermediate goods in each sector,
the variables WRs,t is the time, in quarters, it takes for the sector’s wage contract to be renewed
in period t.
Table 11 reports the coefficient estimates. As in the empirical section, all coefficients are
standardized, and coefficients are shown with standard errors in parentheses. We report results for
several versions of the model. In the simplest version of the model, with neither habit nor hiring
costs, we obtain coefficient estimates of 0.132 and 0.084, respectively for returns and employment.
Those are higher than the reduced-form estimates presented in Section 5, but are in the same
order of magnitude. Notwithstanding, this is strong evidence that the qualitative channels foreseen
in the baseline version of our model can rationalize the empirical evidence well. The monetary
transmission channel in the model goes as follows. In response to an increase in the interest rate
at time t, firms that can adjust wages will be able to reduce their marginal costs. They will then
shelter the negative impact of the shocks on the future sum of discounted profits. The cohort of
firms which cannot adjust wages will experience a larger change in profits, reflected in stock returns.
Next, we extend our model along two dimensions to render the model more realistic and
improve its quantitative performance. We introduce first habit persistence in consumption and
next hiring costs. Habit persistence is introduced for two reasons. First, past macro literature
has already uncovered that this feature improves the model performance in terms of matching
a number of business cycle facts. Second, it allows us to fine tune the response of the stochastic
discount factor of firms, which guides the response to the discount future profits to monetary policy
shocks. Introducing external habits in consumption smooths the marginal weight that investors,
32
and hence firms’ owners, attach to fluctuations in future profits. Hence, it smooths the response of
stock returns to monetary policy shocks. Hiring costs, instead, are chosen since they are a realistic
feature of the labor market and since they tend to delay the response of employment to shocks. In
both those variants of the model regression coefficients become lower, and closer to the empirical
counterparts.
An additional feature which can affect the response of stock values, through its impact on
the stochastic discount factor, is the degree of risk aversion. Raising the coefficient of relative risk
aversion σ from 2 to 2.15 further improves the fit, bringing down the coefficients to 0.025 and
0.056. An increase in risk-aversion implies that investors attach a higher price of risk to future
contingencies and this translates into a more muted response of the stochastic discount factor.
7 Conclusion
The question of the effects of monetary policy in face of nominal rigidities is old, but still highly
relevant. With recent emphasis, in the macro literature, placed on the role of firms’ and agents’
heterogeneity, the question has taken yet a further twist, namely that of the distributional conse-
quences of monetary policy in the face of heterogeneous nominal rigidities.
One of the biggest challenges in answering this question lies in the quest of an exogenous
and accurate measure of nominal rigidities. Recent literature has relied more increasingly on the
use of micro-level data to provide granular metrics of nominal rigidities. Measures of nominal
rigidities have been constructed using price-level data. The latter however are not fully immune to
endogeneity concerns. We go one step further in this direction. By relying on a unique confidential
contract-level dataset for the Italian economy, we are able to construct a fully exogenous measure
of wage rigidity based on a staggered lagged structure of the wage renegotiations. Our research
question looks at the impact of a monetary policy shock on valuations and employment levels when
firms belong to sectors that experience different degrees of rigidity of wages, as set by sectorial
labor unions.
33
We find strong evidence that monetary policy shocks, identified at high frequency, have a sig-
nificant impact on firms that are subject to wage rigidities. The impact is also more significant and
stronger for firms with high labor intensity and low profitability, uncovering the role of combined
dimensions of firms’ heterogeneity. We rationalize our evidence with a dynamic multi-sector gen-
eral equilibrium model in which firms face different degrees of wage rigidity as dictated by union
bargaining and show that the model can rationalize and replicate the empirical evidence quite well.
We conclude by noticing that our empirical design and our model have a more general validity.
Indeed our basic empirical strategy and our laboratory economy can be adapted to study the
response to other common events, such as trade shocks.
34
References
Abowd, John M., Patrick Corbel, and Francis Kramarz, 1999, The Entry and Exit of Workers and
the Growth of Employment: An Analysis of French Establishments, Review of Economics and
Statistics 81, 170–187.
Altavilla, Carlo, Luca Brugnolini, Refet S. Grkaynak, Roberto Motto, and Giuseppe Ragusa, 2019,
Measuring Euro Area Monetary Policy, CEPR Discussion Papers 13759, C.E.P.R. Discussion
Papers.
Altonji, Joseph G., and Paul J. Devereux, 2000, The Extent and Consequences of Downward
Nominal Wage Rigidity, in Research in Labor Economics, 383–431 (Emerald Group Publishing
Limited).
Anderson, Eric, Nir Jaimovich, and Duncan Simester, 2015, Price Stickiness: Empirical Evidence
of the Menu Cost Channel, Review of Economics and Statistics 97, 813–826.
Bahaj, Saleem, Angus Foulis, Gabor Pinter, and Paolo Surico, 2017, Monetary Policy and the
Firm: Some Empirical Evidence .
Barakchian, S Mahdi, and Christopher Crowe, 2013, Monetary Policy Matters: Evidence from New
Shocks Data, Journal of Monetary Economics 60, 950–966.
Belloc, Marianna, Paolo Naticchioni, and Claudia Vittori, 2018, Urban Wage Premia, Cost of
Living, and Collective Bargaining, Working Paper.
Bernanke, Ben S., and Kenneth N. Kuttner, 2005, What Explains the Stock Market’s Reaction to
Federal Reserve Policy?, Journal of Finance 60, 1221–1257.
Bewley, Truman F, 1998, Why Not Cut Pay?, European Economic Review 42, 459–490.
Bjorklund, Maria, Mikael Carlsson, and Oskar Nordstrom Skans, 2019, Fixed-Wage Contracts and
Monetary Non-neutrality, American Economic Journal: Macroeconomics 11, 171–92.
Blinder, Alan S., 1991, Why are Prices Sticky? Preliminary Results from an Interview Study, The
American Economic Review 81, 89–96.
Boeri, Tito, 2015, Perverse Effects of Two-Tier Wage Bargaining Structures, IZA World of Labor
101.
Boeri, Tito, Andrea Ichino, Enrico Moretti, and Johanna Posch, 2017, Unintended Consequences
of Nominal Wage Equality across Regions, Working Paper.
35
Card, David, Jorg Heining, and Patrick Kline, 2013, Workplace Heterogeneity and the Rise of West
German Wage Inequality, The Quarterly Journal of Economics 128, 967–1015.
Card, David, and Dean Hyslop, 1997, Does Inflation “Grease the Wheels of the Labor Market”?,
in Reducing inflation: Motivation and strategy , 71–122 (University of Chicago Press).
Carvalho, Carlos, 2006, Heterogeneity in Price Stickiness and the Real Effects of Monetary Shocks,
Frontiers in Macroeconomics 6.
Cooper, Russell, and Jonathan L. Willis, 2009, The Cost of Cabor Adjustment: Inferences from
the Gap, Review of Economic Dynamics 12, 632–647.
Corsetti, Giancarlo, Joao B. Duarte, and Samuel Mann, 2018, One Money, Many Markets-A Fac-
tor Model Approach to Monetary Policy in the Euro Area with High-Frequency Identification,
Working Paper.
Dell’Aringa, Carlo, 2017, Dai Minimi Tabellari ai Salari di Garanzia, in Lucifora Claudio Del-
lAringa, Carlo, and Tiziano Treu, eds., Produttivita, Diseguaglianze (Il Mulino).
Devicienti, Francesco, Bernardo Fanfani, and Agata Maida, 2016, Collective Bargaining and the
Evolution of Wage Inequality in Italy, Working Paper.
Dickens, William T, Lorenz Goette, Erica L Groshen, Steinar Holden, Julian Messina, Mark E
Schweitzer, Jarkko Turunen, and Melanie E Ward, 2007, How Wages Change: Micro Evidence
from the International Wage Flexibility Project, Journal of Economic Perspectives 21, 195–214.
Dixit, Avinash K., and Joseph E. Stiglitz, 1977, Monopolistic Competition and Optimum Product
Diversity, The American Economic Review 67, 297–308.
Dube, Arindrajit, T William Lester, and Michael Reich, 2010, Minimum Wage Effects Across
State Borders: Estimates Using Contiguous Counties, The review of economics and statistics 92,
945–964.
Fanfani, Bernardo, 2019, The Employment Effects of Collective Bargaining, Working paper, Uni-
versity of Torino.
Fischer, Stanley, 1977, Long-Term Contracts, Rational Expectations, and the Optimal Money
Supply Rule, journal of Political Economy 85, 191–205.
Friesen, Jane, 1997, The Dynamic Demand for Part-time and Full-time Labour, Economica 64,
483–493.
Galı, Jordi, 2015, Monetary Policy, Inflation, and the Business Cycle: an Introduction to the New
Keynesian Framework and its Applications (Princeton University Press).
36
Gertler, Mark, and Simon Gilchrist, 1994, Monetary Policy, Business Cycles, and the Behavior of
Small Manufacturing Firms, The Quarterly Journal of Economics 109, 309–340.
Gertler, Mark, and Peter Karadi, 2013, QE 1 vs. 2 vs. 3. . . : A Framework for Analyzing Large-
Scale Asset Purchases as a Monetary Policy Tool, International Journal of Central Banking 9,
5–53.
Gertler, Mark, and Peter Karadi, 2015, Monetary Policy Surprises, Credit Costs, and Economic
Activity, American Economic Journal: Macroeconomics 7, 44–76.
Gorodnichenko, Yuriy, and Michael Weber, 2016, Are Sticky Prices Costly? Evidence from the
Stock Market, American Economic Review 106, 165–99.
Groshen, Erica, and Mark Schweitzer, 1999, Identifying Inflation’s Grease and Sand Effects in
the Labor Market, in The Costs and Benefits of Price Stability , 273–314 (University of Chicago
Press).
Gurkaynak, Refet S., Brian Sack, and Eric T. Swanson, 2005, Do Actions Speak Louder Than
Words? The Response of Asset Prices to Monetary Policy Actions and Statements, International
Journal of Central Banking 1, 55–93.
Houseman, Susan N, 2001, Why Employers Use Flexible Staffing Arrangements: Evidence from an
Establishment Survey, Industrial and Labor Relations Review 55, 149–170.
Jeenas, Priit, 2018, Monetary Policy Shocks, Financial Structure, and Firm Activity: A Panel
Approach .
Kahn, Shulamit, 1997, Evidence of Nominal Wage Stickiness from Microdata, The American Eco-
nomic Review 87, 993–1008.
Kramarz, Francis, 2001, Rigid Wages: What Have We Learnt from Microeconometric Studies?, in
Advances in Macroeconomic Theory , 194–216 (Springer).
Kudlyak, Marianna, and Juan M Sanchez, 2017, Revisiting the Behavior of Small and Large Firms
during the 2008 Financial Crisis, Journal of Economic Dynamics and Control 77, 48–69.
Manacorda, Marco, 2004, Can the Scala Mobile Explain the Fall and Rise of Earnings Inequality
in Italy? a Semiparametric Analysis, 1977–1993, Journal of Labor Economics 22, 585–613.
McLaughlin, Kenneth J., 1994, Rigid Wages?, Journal of Monetary Economics 34, 383–414.
Mehrotra, Neil, and Nicolas Crouzet, 2017, Small and Large Firms over the Business Cycle, Tech-
nical report.
Moscarini, Giuseppe, and Fabien Postel-Vinay, 2012, The Contribution of Large and Small Em-
ployers to Job Creation in Times of High and Low Unemployment, American Economic Review
102, 2509–39.
37
Nakamura, Emi, and Jon Steinsson, 2008, Five Facts about Prices: A Reevaluation of Menu Cost
Models, The Quarterly Journal of Economics 123, 1415–1464.
Olivei, Giovanni, and Silvana Tenreyro, 2007, The Timing of Monetary Policy Shocks, American
Economic Review 97, 636–663.
Ottonello, Pablo, and Thomas Winberry, 2018, Financial Heterogeneity and the Investment Chan-
nel of Monetary Policy, Working paper, National Bureau of Economic Research, Inc.
Rodnyansky, Alexander, 2018, (un)competitive Devaluations and Firm Dynamics, Technical report.
Ronchi, Maddalena, and Filippo di Mauro, 2017, Wage Bargaining Regimes and Firms’ Adjust-
ments to the Great Recession, Working Paper Series 2051, European Central Bank.
Schmitt-Grohe, Stephanie, and Martin Uribe, 2005, Optimal Fiscal and Monetary Policy in a
Medium-Scale Macroeconomic Model, NBER Macroeconomics Annual 20, 383–425.
Serfling, Matthew, 2016, Firing Costs and Capital Structure Decisions, The Journal of Finance 71,
2239–2286.
Simintzi, Elena, Vikrant Vig, and Paolo Volpin, 2014, Labor Protection and Leverage, The Review
of Financial Studies 28, 561–591.
Smith, Jennifer C., 2000, Nominal Wage Rigidity in the United Kingdom, The Economic Journal
110, 176–195.
Taylor, John B., 1979, Staggered Wage Setting in a Macro Model, The American Economic Review
108–113.
Zbaracki, Mark J., Mark Ritson, Daniel Levy, Shantanu Dutta, and Mark Bergen, 2004, Managerial
and Customer Costs of Price Adjustment: Direct Evidence from Industrial Markets, Review of
Economics and statistics 86, 514–533.
38
8 Figures and Tables
Figure 1The Effect of an Expansionary Monetary Policy Shock on Firm Profit: An ExampleFigure 1 presents the example described in Section 3. The solid black line and the dotted red lineplot a firm’s profit function before and after an expansionary monetary policy shock, respectively.The optimal price level shifts, accordingly, from p∗1 to p∗2. If prices are rigid in the short run, afirm’s profit may benefit or not from the shock, depending on whether its initial price level is toolow (pL) or too high (pH). In the first case, the profit will decrease after the shock, in the secondit will rise.
p∗1pL pHp∗2 Price
Profit
39
Figure 2Collective Agreements by Firm
Panel A#Observations by Number of
Contracts (in ‘000s)
2 4 6
5
10
15
20
Panel BFraction of Wages Paid to
Modal Contract by Percentile
5 10 25 50 75 90 95
0.25
0.5
0.75
1
40
Figure 3Dynamic Relationship between Minimum Wages and Earnings
Figure 3 shows the coefficients ηs, together with the 95% confidence intervals, obtained by estimat-ing equation 5.
t-48 t-42 t-36 t-30 t-24 t-18 t-12 t-6 t t+6 t+12 t+18 t+24 t+30 t+36 t+42t+t+48t+54 t+60 t+66 t+72 t+78 t+84 t+90
0
2
4
t
Table 1Firm Performance and Vacancy Periods
Table A1 presents logit regressions where the dependent variable is a dummyequal to 1 if the proxy for wage rigidity WR (defined in Section 4.3 for details)is equal to zero. ∆Sales is the first difference in the logarithm of revenues;ROA is operating income over lagged total assets. The continuous variables aredemeaned and divided by their standard deviation for ease of interpretation.All regressions include year-quarter dummies and industty dummies using the2-digit SIC code definition. Standard errors, in parentheses, are clustered atthe firm level. ***, **, and * indicate statistically different from zero at the 1%,5%, and 10% level of significance, respectively.
(1) (2) (3)
∆Salest−1 -0.004 0.023(0.035) (0.044)
∆Salest−2 -0.018 0.003(0.037) (0.040)
ROAt−1 -0.017 -0.034(0.063) (0.066)
ROAt−2 -0.000 0.017(0.079) (0.084)
Observations 8,301 5,803 5,465
Year-Quarter FE X X XIndustry FE X X X
41
Table 2Descriptive Statistics
Table 2 has descriptive statistics for the main variables used in the paper. Return2 isthe squared stock return between 19:00 CET and 13:00 CET on announcement dates ofECB key target rates. MP 2 is the square of the change in the 1-year Euro OvernightIndex Average (EONIA) swap rate over the same time horizon. WR is a proxy forthe average number of days left before the renewal of the relevant collective bargainingagreement. (See Section 4.3 for details.) Size is the logarithm of total assets. Leverageis given by non current liabilities plus current liabilities, all divided by total assets.ROA is earnings before interest and debt divided by total assets. Labor intensity istotal monthly wage expenses divided by total assets.
N Mean Median St. Dev. Min Max
Return2 22,709 0.03 0.01 0.09 0.00 0.98MP2 22,709 0.36 0.05 1.00 0.00 7.41MP2 × WR 22,709 1.64 0.08 5.33 0.00 50.72WR 22,709 5.05 5.77 2.01 0.00 7.31Time to Ren. 22,709 375.72 318.12 314.88 0.00 1,489.00Size 22,709 6.76 6.40 2.13 2.84 12.96Leverage 22,709 0.28 0.29 0.16 0.00 0.83ROA 22,709 0.08 0.08 0.07 -0.19 0.31Lab. Int. 22,709 7.27 7.49 1.63 3.33 10.00
42
Table 3Baseline Results
Table 3 presents regressions where the dependent variable is the squared firm’s stock returnbetween 13:00 CET and 19:00 CET. MP 2 is the square of the change in the 1-year Euro OvernightIndex Average (EONIA) swap rate over the same time horizon. WR is a measure of wage rigidity.(See Section 4.3 for details.) Size is the logarithm of total assets. Leverage is the sum of debt incurrent liabilities and debt in non current liabilities, with all divided by total assets. ROA is theratio of EBITDA over total assets. Labor intensity is the ratio of wage expenses (measured in themonth previous to the announcement) over beginning-of-year total assets. Columns 4 through 6also include firm and announcement date fixed effects. All the accounting control variables aremeasured at the beginning of the year. Standard errors, in parentheses, are clustered at the firmlevel. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% level ofsignificance, respectively.
(1) (2) (3) (4) (5) (6)
MP2 × WR 0.017** 0.016** 0.013* 0.022*** 0.022*** 0.020***(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)
MP2 0.042*** 0.044*** 0.047***(0.007) (0.007) (0.008)
WR 0.007 0.008 0.007 -0.001 -0.001 -0.001(0.008) (0.008) (0.008) (0.009) (0.009) (0.009)
Size -0.017 -0.017 -0.050 -0.064(0.019) (0.019) (0.065) (0.062)
Leverage 0.029** 0.029** 0.026 0.028(0.014) (0.014) (0.020) (0.020)
ROA -0.045*** -0.045*** -0.003 -0.003(0.011) (0.011) (0.019) (0.019)
Lab. Int. 0.027* 0.027* 0.011 0.010(0.016) (0.016) (0.034) (0.034)
MP2 × Size -0.029*** -0.028***(0.010) (0.010)
MP2 × Lev. 0.009 0.010(0.011) (0.010)
MP2 × ROA -0.008 -0.006(0.009) (0.009)
MP2 × Lab. 0.002 0.001(0.010) (0.010)
Observations 22,709 22,709 22,709 22,709 22,709 22,709R2 0.002 0.006 0.007 0.130 0.130 0.131
Controls X X X XControls×MP X XTime FE X X XFirm FE X X X
43
Table 4Robustness Tests: Industry Effects and Swap Horizons
Table 4 presents regressions where the dependent variable is the squared firm’s stockreturn between 13:00 CET and 19:00 CET. MP 2 is the square of the change in the1-year Euro Overnight Index Average (EONIA) swap rate over the same time hori-zon, except in columns 2 and 3, where we use the 6-month and the 3-month swap,respectively. WR is a measure of wage rigidity. (See Section 4.3 for details.) Allregressions include, as control variables, size, leverage, ROA and labor intensity, bothas standalone variables and interacted with MP 2, all measured at the beginning ofthe year. Size is the logarithm of total assets. Leverage is the sum of debt in currentliabilities and debt in non current liabilities, with all divided by total assets. ROA isthe ratio of EBITDA over total assets. Labor intensity is the ratio of wage expensesover total assets. All the regressions include firm and announcement date fixed effects.In columns 4 through 6, regressions also include industry dummies interacted withMP 2. In columns 4, 5 and 6, industries are defined using the 1, 2 or 3-digit NACEclassification, respectively. Standard errors, in parentheses, are clustered at the firmlevel. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10%level of significance, respectively.
Using Different Measuresof MP Shocks:︷ ︸︸ ︷
Controlling forIndustry FE×Shock:︷ ︸︸ ︷
1-YearSwap
6-MoSwap
3-MoSwap
1-DigitNACE
2-DigitNACE
3-DigitNACE
WR -0.001 -0.002 -0.002 -0.002 -0.006 -0.008(0.009) (0.009) (0.009) (0.009) (0.009) (0.010)
MP2 × WR 0.020*** 0.021*** 0.023** 0.017*** 0.017*** 0.019***(0.007) (0.007) (0.010) (0.006) (0.006) (0.007)
Observations 22,709 22,709 22,709 22,709 19,991 16,603R2 0.131 0.131 0.131 0.132 0.135 0.143
Controls X X X X X XControls×MP2 X X X X X XTime FE X X X X X XFirm FE X X X X X XInd. FE×MP2 X X X
44
Table 5Robustness Tests: Different Windows around Announcements
Table 5 presents regressions where the dependent variable is the squared firm’s stock returnover several windows. MP 2 is the square of the change in the median quote of the 1-yearEuro Overnight Index Average (EONIA) swap rate from the window 13:25-13:35 before thepress release to the median quote in the window 15:40-15:50 after the press conference. Thedependent variable is the squared stock return of the same horizon in column 1, between13:00 and 19:00 in column 2 and over the entire day in column 3. WR is a measure of wagerigidity. (See Section 4.3 for details.) All regressions include, as control variables, size,leverage, ROA and labor intensity, both as standalone variables and interacted with MP 2,all measured at the beginning of the year. Size is the logarithm of total assets. Leverage isthe sum of debt in current liabilities and debt in non current liabilities, with all divided bytotal assets. ROA is the ratio of EBITDA over total assets. Labor intensity is the ratio ofwage expenses over total assets. All the regressions include firm and announcement datefixed effects. In columns 4 through 6, regressions also include industry dummies interactedwith MP 2. In columns 4, 5 and 6, industries are defined using the 1, 2 or 3-digit NACEclassification, respectively. Standard errors, in parentheses, are clustered at the firm level.***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% level ofsignificance, respectively.
Window of Dep. Variable:Altavilla et al.
(2019)Corsetti et al.
(2019)Daily
(1) (2) (3)
MP2 × WR 0.012** 0.017*** 0.015**(0.006) (0.006) (0.007)
WR 0.021*** -0.002 -0.002(0.008) (0.009) (0.008)
Observations 22,709 22,709 22,287R2 0.078 0.131 0.094
Controls X X XControls×MP2 X X XTime FE X X XFirm FE X X X
45
Table 6The Effect of Labor Intensity
Table 6 presents regressions where the dependent variable is the squared firm’s stockreturn between 13:00 CET and 19:00 CET. MP 2 is the square of the change in the 1-year Euro Overnight Index Average (EONIA) swap rate over the same time horizon.WR is a measure of wage rigidity. (See Section 4.3 for details.) All regressions in-clude as control variables size, leverage, ROA and labor intensity, both as standalonevariables and interacted with MP 2, all measured at the beginning of the year. Sizeis the logarithm of total assets. Leverage is the sum of debt in current liabilities anddebt in non current liabilities, with all divided by total assets. ROA is the ratio ofEBITDA over total assets. Labor intensity is the ratio of wage expenses over total as-sets. All the regressions include firm and announcement date fixed effects. Columns1 and 2 include firms that have labor intensity below and above the sample median,respectively. In columns 3 and 4 the proxy for labor intensity is given by total daysworked divided by total assets. The numerators of both proxies are measured in themonth previous to the announcement date, whereas total assets are measured at thebeginning of the year. Standard errors, in parentheses, are clustered at the firm level.***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% levelof significance, respectively.
Sorting by:Wages / Assets Days Worked / Assets
Low High Low High(1) (2) (3) (4)
MP2 × WR 0.014 0.026** 0.015* 0.025**(0.009) (0.012) (0.009) (0.012)
WR -0.010 0.003 -0.014 0.009(0.012) (0.012) (0.012) (0.012)
Observations 11,337 11,360 11,371 11,333R2 0.191 0.108 0.190 0.110
Controls X X X XControls×MP X X X XTime FE X X X XFirm FE X X X X
46
Table 7Firm’s Performance and Business Cycle
Table 7 presents regressions where the dependent variable is the squared firm’s stockreturn between 13:00 CET and 19:00 CET. MP 2 is the square of the change in the 1-year Euro Overnight Index Average (EONIA) swap rate over the same time horizon.WR is a measure of wage rigidity. (See Section 4.3 for details.) All regressions includeas control variables size, leverage, ROA and labor intensity, both as standalone vari-ables and interacted with MP 2, all measured at the beginning of the year. Size is thelogarithm of total assets. Leverage is the sum of debt in current liabilities and debt innon current liabilities, with all divided by total assets. ROA is the ratio of EBITDAover total assets. Labor intensity is the ratio of wage expenses over total assets. Allthe regressions include firm and announcement date fixed effects. Columns 1 and 2include firms that have ROA below and above the sample median, respectively. Incolumns 3 and 4 the sorting variable is the quarterly GDP growth. Standard errors,in parentheses, are clustered at the firm level. ***, **, and * indicate statisticallydifferent from zero at the 1%, 5%, and 10% level of significance, respectively.
Sorting by:ROA GDP Growth
Low High Low High(1) (2) (3) (4)
MP2 × WR 0.026** 0.014 0.024*** 0.009(0.011) (0.010) (0.008) (0.010)
WR -0.014 0.002 0.005 -0.008(0.012) (0.012) (0.013) (0.013)
Observations 10,480 12,222 11,431 11,269R2 0.149 0.142 0.139 0.138
Controls X X X XControls×MP X X X XTime FE X X X XFirm FE X X X X
47
Table 8The Effects of Part-Time and Fixed-Term Contracts
Table 8 presents regressions where the dependent variable is the squared firm’s stockreturn between 13:00 CET and 19:00 CET. MP 2 is the square of the change in the 1-year Euro Overnight Index Average (EONIA) swap rate over the same time horizon.WR is a measure of wage rigidity. (See Section 4.3 for details.) All regressions in-clude as control variables size, leverage, ROA and labor intensity, both as standalonevariables and interacted with MP 2, all measured at the beginning of the year. Sizeis the logarithm of total assets. Leverage is the sum of debt in current liabilities anddebt in non current liabilities, with all divided by total assets. ROA is the ratio ofEBITDA over total assets. Labor intensity is the ratio of wage expenses over total as-sets. All the regressions include firm and announcement date fixed effects. Columns1 and 2 include firms that have a fraction of workers with part-time contracts belowand above the sample median, respectively. In columns 3 and 4 the sorting variableis the fraction of workers with fixed-term contracts. Standard errors, in parentheses,are clustered at the firm level. ***, **, and * indicate statistically different from zeroat the 1%, 5%, and 10% level of significance, respectively.
Sorting by Fraction of Workers with:Part-Time Contracts Fixed-Term Contracts
Low High Low High(1) (2) (3) (4)
MP2 × WR 0.021*** 0.005 0.024*** 0.017*(0.008) (0.012) (0.008) (0.010)
WR 0.006 -0.018 -0.003 -0.000(0.012) (0.013) (0.012) (0.012)
Observations 11,466 11,243 11,382 11,327R2 0.148 0.130 0.142 0.133
Controls X X X XControls×MP X X X XTime FE X X X XFirm FE X X X X
48
Table 9Labor Outcomes
Table 9 presents regressions where the dependent variables are several employment outcomes,measured at the quarterly horizon. MP 2 is the square of the weighted sum of changes inthe 1-year Euro Overnight Index Average (EONIA) swap rate over the quarter. WR isa measure of wage rigidity. (See Section 4.3 for details.) All regressions include as controlvariables size, leverage, ROA and labor intensity, both as standalone variables and interactedwith MP 2, all measured at the beginning of the year. Size is the logarithm of total assets.Leverage is the sum of debt in current liabilities and debt in non current liabilities, with alldivided by total assets. ROA is the ratio of EBITDA over total assets. Labor intensity is theratio of wage expenses over total assets. All the regressions include firm and announcementdate fixed effects. The dependent variables are the symmetric growth rates of: total wagepayments (column 1), total days worked (column 2), total number of employees (column 3),total number of employees with at least 20 days worked in the month (column 4). Standarderrors, in parentheses, are clustered at the firm level. ***, **, and * indicate statisticallydifferent from zero at the 1%, 5%, and 10% level of significance, respectively.
Dep. Var. ∆Pay ∆Days Worked ∆Employees∆Full TimeEmployees
(1) (2) (3) (4)
MP2 × WR 0.038*** 0.030* 0.037** 0.014(0.013) (0.016) (0.017) (0.015)
WR 0.002 0.008 0.010 -0.001(0.010) (0.013) (0.012) (0.013)
Observations 10,928 10,928 10,928 10,928R2 0.418 0.239 0.227 0.231
Controls X X X XControls×MP2 X X X XTime FE X X X XFirm FE X X X X
49
Table 10Model Calibration
Table 10 presents, for each parameter of the model presented in Section 6, the value chosen for thecalibration with the relevant source.
Value Description Source
β 0.99 Discount factor Standard
bW 1.5 Response coefficient in mon. pol. rule Standard
σ 2 Coefficient of relative risk aversion Standard
η 1.17 Inverse Frisch labor elasticity Fernandez-Villaverde et al. (2010)
εp 10 Elasticity of substitution of goods Fernandez-Villaverde et al. (2010)
εw 10 Elasticity of substitution of labor services Fernandez-Villaverde et al. (2010)
ρi 0.77 Smoothing parameter in mon. pol. rule Fernandez-Villaverde et al. (2010)
fs 1/S Sector shares Avg. renegotation time in data
S 8 Number of sectors Avg. renegotation time in data
σi 0.0043 Volatility of monetary policy shock Gorodnichenko and Weber (2016)
h 0.815 Internal consumption habit Gertler and Karadi (2013)
ξ 2.015 Hiring cost parameter Cooper and Willis (2009) (avg.)
Table 11Regression Coefficients Estimated on Artificial Data
Table 11 presents coefficients estimated on an artificial dataset generated by the model describedin Section 6, with parameter values calibrated using the values indicated in Table 10. In column1 the dependent variable is the firm stock return squared. In column 2 it is the symmetric growthrate of employment squared. The regressors are the wage rigidity proxy, the monetary policy shockand an interaction term of the two. Only the coefficients associated to the latter regressor areshowed. In the first row the coefficients are estimated on a simplified version of the model that hasneither an habit component in the utility function, nor hiring costs. The second row presents datagenerated from the fully specified model. In the third row the model is identical but the relativerisk aversion parameter is increased from 2 to 2.15.
Specification Stock Return Employment Growth
Baseline 0.132*** 0.084***
(0.011) (0.012)
...plus habit and hiring costs 0.033*** 0.071***
(0.012) (0.012)
...plus habit and hiring costs, 0.025** 0.056***
relative risk aversion = 2.15 (0.012) (0.012)
50
A Appendix
In this Appendix we include robustness tests omitted from the main text for brevity. In Table A1
we test whether vacancy periods are systematically predicted by the performance of the firms that
employ such workers. On Compustat quarterly data, we estimate the following logit model:
Pr (WRi,t = 0|Xi,t−1, Xi,t−2, δt, ηj) = G(β′Xi,t−1 + γ′Xi,t−2 + δt + ηj + εi,t
)(A.1)
where G is the cumulative logistic function, and i, t and j index firms, year-quarters and industries,
respectively. δt is a vector of time dummies, and ηj a vector of industry fixed effects. The vector
X includes predictors of vacancy periods, that are related to firm performance. We include the
change in log-revenues, and ROA, a measure of profitability (measured as operating profit over
lagged total assets). We include two lags of each predictor, and demean and standardize each
regressor. Table A1 shows that, no matter whether we include only sale growth (column 1), only
profitability (column 2), or both (column 3) as regressors, firm performance does not appear to
predict vacancy periods. Hence, we are reassured that the censoring of our proxy is appropriate
and unlikely to mask some relevant omitted variables.
We also test for the presence of any pre-trend in stock market returns that could anticipate the
monetary policy shocks. More specifically, we replace our dependent variable with the stock return
between 1pm and 7pm on the day prior to each monetary policy announcement. If the correlations
found so far were spurious, we could in principle obtain similar estimates even in days in which
monetary policy announcements do not take place. On the other hand, a small and insignificant
coefficient on the interaction term of interest would suggest that our results are indeed driven by
heterogeneity in the response to monetary policy shocks.
In Table A2 we first replicate our baseline test of Table 3, focusing for brevity on the most
conservative specification. In column 1 we find a coefficient on the interaction term that is small
and insignificant, equal to -0.003 (t-statistic=0.38). We also replicate our analysis of Table 6, where
we sort firms according to their degree of labor intensity. This test is crucial because it provides
51
strong evidence that the labor channel is indeed the driving force of our results. As columns 2
through 5 show, the coefficient of interest remains small and insignificant in all the subsamples
and, if anything, is more negative in the high labor intensity subgroups.
In Table A3, we perform a similar “placebo test” for the regressions with labor market outcomes
presented in Table 9. Here we use as dependent variables the one quarter-lags of those used in our
main tests. As expected, we do not find any significant relationship between changes in employment
and shocks occurring in the following quarters.
52
Table A1Firm Performance and Vacancy Periods
Table A1 presents logit regressions where the dependent variable is a dummyequal to 1 if the proxy for wage rigidity WR (defined in Section 4.3 for details)is equal to zero. ∆Sales is the first difference in the logarithm of revenues;ROA is operating income over lagged total assets. The continuous variables aredemeaned and divided by their standard deviation for ease of interpretation.All regressions include year-quarter dummies and industty dummies using the2-digit SIC code definition. Standard errors, in parentheses, are clustered atthe firm level. ***, **, and * indicate statistically different from zero at the 1%,5%, and 10% level of significance, respectively.
(1) (2) (3)
∆Salest−1 -0.004 0.023
(0.035) (0.044)
∆Salest−2 -0.018 0.003
(0.037) (0.040)
ROAt−1 -0.017 -0.034
(0.063) (0.066)
ROAt−2 -0.000 0.017
(0.079) (0.084)
Observations 8,301 5,803 5,465
Year-Quarter FE X X X
Industry FE X X X
53
Table A2Placebo Test
Table A2 presents regressions where the dependent variable is the squared firm’s stock returnbetween 13:00 CET and 19:00 CET on the days prior to the monetary policy announcementdates. MP 2 is the square of the change in the 1-year Euro Overnight Index Average (EONIA)swap rate over the same time horizon. WR is a measure of wage rigidity. (See Section 4.3 fordetails.) All regressions include as control variables size, leverage, ROA and labor intensity,both as standalone variables and interacted with MP 2, all measured at the beginning of theyear. Size is the logarithm of total assets. Leverage is the sum of debt in current liabilitiesand debt in non current liabilities, with all divided by total assets. ROA is the ratio ofEBITDA over total assets. Labor intensity is the ratio of wage expenses over total assets.All the regressions include firm and announcement date fixed effects. Column 1 includes thefull sample. Columns 2 and 3 include firms that have labor intensity below and above thesample median, respectively. In columns 4 and 5 the proxy for labor intensity is given bytotal days worked divided by total assets. The numerators of both proxies are measuredin the month previous to the announcement date, whereas total assets are measured at thebeginning of the year. Standard errors, in parentheses, are clustered at the firm level. ***, **,and * indicate statistically different from zero at the 1%, 5%, and 10% level of significance,respectively.
Sorting by:︷ ︸︸ ︷Baseline Wages / Assets Days Worked / Assets
Low High Low High
(1) (2) (3) (4) (5)
MP2 × WR -0.004 -0.002 -0.009 -0.003 -0.011
(0.008) (0.010) (0.015) (0.010) (0.012)
WR 0.008 0.014 -0.005 0.019* -0.005
(0.008) (0.012) (0.011) (0.011) (0.011)
Observations 22,301 10,232 12,064 11,136 11,156
R2 0.088 0.106 0.094 0.099 0.101
Controls X X X X X
Controls×MP X X X X X
Time FE X X X X X
Firm FE X X X X X
54
Table A3Labor Outcomes: Placebo test
Table A3 presents regressions where the dependent variables are several employment out-comes, measured at the quarterly horizon and lagged one quarter. MP 2 is the square ofthe weighted sum of changes in the 1-year Euro Overnight Index Average (EONIA) swaprate over the quarter. WR is a measure of wage rigidity. (See Section 4.3 for details.)All regressions include as control variables size, leverage, ROA and labor intensity, both asstandalone variables and interacted with MP 2, all measured at the beginning of the year.Size is the logarithm of total assets. Leverage is the sum of debt in current liabilities anddebt in non current liabilities, with all divided by total assets. ROA is the ratio of EBITDAover total assets. Labor intensity is the ratio of wage expenses over total assets. All theregressions include firm and announcement date fixed effects. The dependent variables arethe symmetric growth rates of: total wage payments (column 1), total days worked (column2), total number of employees (column 3), total number of employees with at least 20 daysworked in the month (column 4). Standard errors, in parentheses, are clustered at the firmlevel. ***, **, and * indicate statistically different from zero at the 1%, 5%, and 10% levelof significance, respectively.
Dep. Var. ∆Payt−1∆Days
Workedt−1∆Employeest−1
∆Full TimeEmployeest−1
(1) (2) (3) (4)
MP2 × WR -0.022 -0.004 -0.017 -0.017
(0.015) (0.018) (0.023) (0.023)
WR 0.001 0.007 0.010 0.003
(0.009) (0.012) (0.012) (0.012)
Observations 12,095 12,095 12,095 12,095
R2 0.402 0.229 0.216 0.225
Controls X X X X
Controls×MP2 X X X X
Time FE X X X X
Firm FE X X X X
55